Integration with pytest test framework

Please mind that integration of twister with pytest is still work in progress. Not every platform type is supported in pytest (yet). If you find any issue with the integration or have an idea for an improvement, please, let us know about it and open a GitHub issue/enhancement.

Introduction

Pytest is a python framework that “makes it easy to write small, readable tests, and can scale to support complex functional testing for applications and libraries” (https://docs.pytest.org/en/7.3.x/). Python is known for its free libraries and ease of using it for scripting. In addition, pytest utilizes the concept of plugins and fixtures, increasing its expendability and reusability. A pytest plugin pytest-twister-harness was introduced to provide an integration between pytest and twister, allowing Zephyr’s community to utilize pytest functionality with keeping twister as the main framework.

Integration with twister

By default, there is nothing to be done to enable pytest support in twister. The plugin is developed as a part of Zephyr’s tree. To enable install-less operation, twister first extends PYTHONPATH with path to this plugin, and then during pytest call, it appends the command with -p twister_harness.plugin argument. If one prefers to use the installed version of the plugin, they must add --allow-installed-plugin flag to twister’s call.

Pytest-based test suites are discovered the same way as other twister tests, i.e., by a presence of test/sample.yaml. Inside, a keyword harness tells twister how to handle a given test. In the case of harness: pytest, most of twister workflow (test suites discovery, parallelization, building and reporting) remains the same as for other harnesses. The change happens during the execution step. The below picture presents a simplified overview of the integration.

If harness: pytest is used, twister delegates the test execution to pytest, by calling it as a subprocess. Required parameters (such as build directory, device to be used, etc.) are passed through a CLI command. When pytest is done, twister looks for a pytest report (results.xml) and sets the test result accordingly.

How to create a pytest test

An example folder containing a pytest test, application source code and Twister configuration .yaml file can look like the following:

test_foo/
├─── pytest/
│    └─── test_foo.py
├─── src/
│    └─── main.c
├─── CMakeList.txt
├─── prj.conf
└─── testcase.yaml

An example of a pytest test is given at samples/subsys/testsuite/pytest/shell/pytest/test_shell.py. Using the configuration provided in the testcase.yaml file, Twister builds the application from src and then, if the .yaml file contains a harness: pytest entry, it calls pytest in a separate subprocess. A sample configuration file may look like this:

tests:
   some.foo.test:
      harness: pytest
      tags: foo

By default, pytest tries to look for tests in a pytest directory located next to a directory with binary sources. A keyword pytest_root placed under harness_config section in .yaml file can be used to point to other files, directories or subtests (more info here).

Pytest scans the given locations looking for tests, following its default discovery rules.

Passing extra arguments

There are two ways for passing extra arguments to the called pytest subprocess:

From .yaml file, using pytest_args placed under harness_config section - more info here.
Through Twister command line interface as --pytest-args argument. This can be particularly useful when one wants to select a specific testcase from a test suite. For instance, one can use a command:
```
$ ./scripts/twister --platform native_sim -T samples/subsys/testsuite/pytest/shell \
-s samples/subsys/testsuite/pytest/shell/sample.pytest.shell \
--pytest-args='-k test_shell_print_version'
```
The command line arguments will extend those from the .yaml file. If the same argument is present in both places, the one from the command line will take precedence.

Fixtures

dut

Give access to a DeviceAdapter type object, that represents Device Under Test. This fixture is the core of pytest harness plugin. It is required to launch DUT (initialize logging, flash device, connect serial etc). This fixture yields a device prepared according to the requested type (native, qemu, hardware, etc.). All types of devices share the same API. This allows for writing tests which are device-type-agnostic. Scope of this fixture is determined by the pytest_dut_scope keyword placed under harness_config section (more info here).

from twister_harness import DeviceAdapter

def test_sample(dut: DeviceAdapter):
   dut.readlines_until('Hello world')

shell

Provide a Shell class object with methods used to interact with shell application. It calls wait_for_promt method, to not start scenario until DUT is ready. The shell fixture calls dut fixture, hence has access to all its methods. The shell fixture adds methods optimized for interactions with a shell. It can be used instead of dut for tests. Scope of this fixture is determined by the pytest_dut_scope keyword placed under harness_config section (more info here).

from twister_harness import Shell

def test_shell(shell: Shell):
   shell.exec_command('help')

mcumgr

Sample fixture to wrap mcumgr command-line tool used to manage remote devices. More information about MCUmgr can be found here MCUmgr.

Note

This fixture requires the mcumgr available in the system PATH

Only selected functionality of MCUmgr is wrapped by this fixture. For example, here is a test with a fixture mcumgr

from twister_harness import DeviceAdapter, Shell, McuMgr

def test_upgrade(dut: DeviceAdapter, shell: Shell, mcumgr: McuMgr):
   # free the serial port for mcumgr
   dut.disconnect()
   # upload the signed image
   mcumgr.image_upload('path/to/zephyr.signed.bin')
   # obtain the hash of uploaded image from the device
   second_hash = mcumgr.get_hash_to_test()
   # test a new upgrade image
   mcumgr.image_test(second_hash)
   # reset the device remotely
   mcumgr.reset_device()
   # continue test scenario, check version etc.

unlaunched_dut

Similar to the dut fixture, but it does not initialize the device. It can be used when a finer control over the build process is needed. It becomes responsibility of the test to initialize the device.

from twister_harness import DeviceAdapter

def test_sample(unlaunched_dut: DeviceAdapter):
   unlaunched_dut.launch()
   unlaunched_dut.readlines_until('Hello world')

Classes

DeviceAdapter

class twister_harness.DeviceAdapter(device_config: DeviceConfig)

This class defines a common interface for all device types (hardware, simulator, QEMU) used in tests to gathering device output and send data to it.

launch() → None: Start by closing previously running application (no effect if not needed). Then, flash and run test application. Finally, start an internal reader thread capturing an output from a device.

connect(retry_s: int = 0) → None: Connect to device - allow for output gathering.

readline(timeout: float | None = None, print_output: bool = True) → str: Read line from device output. If timeout is not provided, then use base_timeout.

readlines(print_output: bool = True) → list[str]: Read all available output lines produced by device from internal buffer.

readlines_until(regex: str | None = None, num_of_lines: int | None = None, timeout: float | None = None, print_output: bool = True) → list[str]

Read available output lines produced by device from internal buffer until following conditions:

If regex is provided - read until regex regex is found in read line (or until timeout).
If num_of_lines is provided - read until number of read lines is equal to num_of_lines (or until timeout).
If none of above is provided - return immediately lines collected so far in internal buffer.

If timeout is not provided, then use base_timeout.

write(data: bytes) → None: Write data bytes to device.

disconnect() → None: Disconnect device - block output gathering.

close() → None: Disconnect, close device and close reader thread.

Shell

class twister_harness.Shell(device: DeviceAdapter, prompt: str = 'uart:~$', timeout: float | None = None)

Helper class that provides methods used to interact with shell application.

exec_command(command: str, timeout: float | None = None, print_output: bool = True) → list[str]: Send shell command to a device and return response. Passed command is extended by double enter sings - first one to execute this command on a device, second one to receive next prompt what is a signal that execution was finished. Method returns printout of the executed command.

wait_for_prompt(timeout: float | None = None) → bool: Send every 0.5 second “enter” command to the device until shell prompt statement will occur (return True) or timeout will be exceeded (return False).

Examples of pytest tests in the Zephyr project

Pytest shell application testing
MCUmgr tests - tests/boot/with_mcumgr
LwM2M tests - tests/net/lib/lwm2m/interop
GDB stub tests - tests/subsys/debug/gdbstub

FAQ

How to flash/run application only once per pytest session?

dut is a fixture responsible for flashing/running application. By default, its scope is set as function. This can be changed by adding to .yaml file pytest_dut_scope keyword placed under harness_config section:
harness: pytest
harness_config:
   pytest_dut_scope: session
More info can be found here.

How to run only one particular test from a python file?

This can be achieved in several ways. In .yaml file it can be added using a pytest_root entry placed under harness_config with list of tests which should be run:
harness: pytest
harness_config:
   pytest_root:
      - "pytest/test_shell.py::test_shell_print_help"
Particular tests can be also chosen by pytest -k option (more info about pytest keyword filter can be found here ). It can be applied by adding -k filter in pytest_args in .yaml file:
harness: pytest
harness_config:
   pytest_args:
      - "-k test_shell_print_help"
or by adding it to Twister command overriding parameters from the .yaml file:
$ ./scripts/twister ... --pytest-args='-k test_shell_print_help'

How to get information about used device type in test?

This can be taken from dut fixture (which represents DeviceAdapter object):
device_type: str = dut.device_config.type
if device_type == 'hardware':
   ...
elif device_type == 'native':
   ...

How to rerun locally pytest tests without rebuilding application by Twister?

This can be achieved by running Twister once again with --test-only argument added to Twister command. Another way is running Twister with highest verbosity level (-vv) and then copy-pasting from logs command dedicated for spawning pytest (log started by Running pytest command: ...).

Is this possible to run pytest tests in parallel?

Basically pytest-harness-plugin wasn’t written with intention of running pytest tests in parallel. Especially those one dedicated for hardware. There was assumption that parallelization of tests is made by Twister, and it is responsible for managing available sources (jobs and hardwares). If anyone is interested in doing this for some reasons (for example via pytest-xdist plugin) they do so at their own risk.

Limitations

Not every platform type is supported in the plugin (yet).